Carol Corrado, Jonathan Haskel and Cecilia Jona-Lasinio
- Date Posted:
- Is Database:
- Database
AI investments have not translated into sustained TFP growth post-financial crisis, despite significant unmeasured investment in AI-related intangible assets. @CarolCorrado
Carol Corrado on the potential impact of AI on productivity, basically fails to find support for the J curve camp, key quote, "... Our main finding is that, on our data at least, there is indeed plenty of unmeasured investment but little sign of a ‘J-curve’ effect on TFP growth. The upward ‘swoosh’ of the effects of investments whose returns are long-lasting just is not there..."
Core of paper, "...We have set out a framework to illustrate this effect and used a cross-country-industry-year data set for the US and European economies to examine it. To look for unmeasured investment, we have used the CHS approach, which brings into national accounts unmeasured investment in intangible assets which are likely complementary to artificial intelligence, such as design, training, and business process re-engineering. We have also seen that at least some artificial intelligence investment is likely in software investment and is thus already counted. Thus we have harmonized the deflation of software and hardware investment to facilitate comparison across different countries. At least on these data, we do not find much support for the J-curve view. There is, indeed, plenty of unmeasured investment, but the trend in such investment does not seem to be sufficient to give an effect. In particular, the high depreciation rates mean that the missing capital payments which bias TFP growth up follow quickly after a burst of intangible investment, meaning that the upward-sloping part of the J curve, at least on our data, appears very quickly, too quickly to account for a sustained TFP growth slowdown following the financial crisis. That said, we are in the early stages of measuring AI, and since much of it is taking place within firms on their own account, the detection of such investment is extremely difficult..."
The evidence
"...Figure 8 shows shares of tangible and intangible investment across countries from 1997 to 2017. As the graph shows, intangible investment is generally trended upwards and the US invests considerably more intangible investment then the EU. This immediately suggested accounting for intangible investment might potentially be important...."
"...As we have discussed, national accounts do indeed measure intangible investment. Figure 9 plots national accounts intangible investment and the additional investment suggested by CHS set out in Table 2. As the graph shows, the non-national accounts intangible share is higher than the national accounts intangible share, reflecting the fact that expenditure on items such as training and design are large. However, the trend is towards relatively more intensive national accounts intangibles measurement..."
"...Finally, Figure 10 shows growth in real investment for tangibles, intangibles, and software. Notice that, particularly in the US, software spending grows particularly strongly at the end of the period. If artificial intelligence is included in such spending this is suggestive. The position seems much more volatile in Europe..."
"...To start our examination of total factor productivity,Figure 11 shows growth accounting for the EU and US for the years 1997-2007 and 2010-17 (we omit the recession years of 2008 and 2009). As the top panel shows, labour productivity and TFP were growing at a healthy rate in the run up to the financial crisis. The contributions of both tangible and intangible capital deepening were higher in the US than in Europe, and their contributions exceeded those of labour composition. In the period following the financial crisis, the situation changed. As is well known, labour productivity growth slowed substantially by around 0.5 percentage points per annum (pppa) in Europe and 2pppa in the US. US TFP growth also slowed: if anything European TFP growth was around the same. Intangible capital deepening slowed very strongly in the US, and somewhat in Europe...."
"... Figure 12 shows the biases to measured TFP growth for the EU and the US respectively: these are the terms on the right of equation (4), second line. A positive number indicates that measured TFP growth is too big. Recall that the ‘J curve’ hypothesis is that measured TFP growth after the financial crisis is too small and then rises, suggesting that after the financial crisis the bias line should dip and then perhaps rise. At least for the EU, there is not much support for this hypothesis. Note first that the biases are small, at most around 0.2 per cent per annum, but less than this for most of the sample. On average, in the EU the bias is slightly positive but seems to show no particular trend. Turning to the US, there is a hint of a J-curve effect after the mid-2000s. The bias, which was almost 0.4 per cent per annum fell steadily to around -0.2 per cent per annum in 2011, with the spike in the financial crisis years which presumably reflects mismeasured utilization and the like. Since then, the bias has been moving back towards being positive. This then suggests that the pre-crisis TFP growth slow-down, which has been noted by a number of authors, may be somewhat overstated. That said, the effects do not appear to be all that large...."
"...What is the intuition behind this apparent finding that the positive ‘swoosh’ of the ‘J’ appears quite quickly?We can get some insight into this from Figure 13, which shows the different components of the mismeasurement on the right-hand side of equation (4). These are the capitalization effect, the errors-in-shares effect, and the mismeasured capital payments effect. What is notable is that the mismeasured capital payments effect comes in very strongly and quite quickly. The intuition here would seem to be the following. As set out in equation (3), intangibles depreciate quite quickly. That means that a burst in intangible investment rapidly builds up the intangible stock, and the rental price on that stock is relatively large (the per-period rental price of capital has to be large for capital that depreciates to compensate the capital owner for renting out an asset whose value will fall quickly). As a result, the missing rental payments effect comes in very quickly following even unmeasured intangible investment. That means that the second half of the ‘J’ appears rather quickly after the initial dip. All this suggests that investments in artificial intelligence will have a more substantial ‘J effect’ the more they are mismeasured and the lower their depreciation rate. Now, it might be that current AI investments are not substantially mismeasured for the simple reason that we might still be too early in the AI investment cycle. As the earlier graphs showed, the ‘second wave’ of AI might be comparatively recent and so it might be too early to pick up the mismeasurement involved. It might also be that we have mismeasured depreciation rates. At the moment, in our system AI-related investments are given the high depreciation rates the literature has found are appropriate to intangibles. But it may, of course, be the case that artificial intelligence investment has a much lower depreciation rate...."
Carol Corrado, Jonathan Haskel and Cecilia Jona-Lasinio, "Artificial intelligence and productivity: an intangible assets approach," Oxford Review Of Economic Policy, 2021, https://academic.oup.com/oxrep/article/37/3/435/6374681